1. Field of the Invention
The present invention relates to semiconductor non-volatile memory technology and more particularly to a structure of and a method for producing a contact-less array of self-aligned, triple polysilicon, source-side injection, flash memory cells.
2. Description of Related Art
FIGS. 1A-1C show different perspectives of a contact-less array of triple-polysilicon, source side injection, flash EPROM cells disclosed by Ma et al. in U.S. Pat. No. 5,280,446 issued Jan. 18, 1994, and incorporated herein by reference.
In FIG. 1A, each cell includes a drain diffusion 40, a source diffusion 50, a floating gate 10 (first layer poly), a control gate 20 (second layer poly), and a select gate 30 (third layer poly). The floating gate 10 and the control gate 20 extend over a first portion L1 of the channel region L. The source diffusion 50 is laterally spaced a distance L2 from the floating gate 10. L2 forms a second portion of the channel region L. The drain diffusion 40 is self-aligned with the stack of floating gate 10 and control gate 20. This cell structure is commonly referred to as xe2x80x9csplit gatexe2x80x9d because it merges two serially connected transistors (i.e., the select gate transistor and the floating gate transistor) into a single memory cell. The select gate 30 extends in a direction which is perpendicular to a drain extension, and runs over the drain diffusion 40, the control gate 20, the portion L2 of the channel region L, and the source diffusion 50 of every cell in a row of such cells.
A layout diagram of two rows of memory cells, each row corresponding to the cross section view of FIG. 1A, is shown in FIG. 1B. The floating gates are shown as the cross hatched regions 10; the drain diffusions are connected together forming a column 40 (drain bitline); the source diffusions are connected together forming another bitline column 50 (source bitline); the control gates are connected together forming yet another column 20 (polysilicon line); and the select gates are connected together forming a row 30 (wordline) perpendicular to the columns.
The drain and source diffusion bitlines are strapped with metal lines (not shown) to minimize the resistance associated with the diffusion bitlines. This is necessary in order to achieve the desired read and programming characteristics. Contacts are used to strap the diffusion bitlines with metal (e.g., one contact may be used every 64 or 128 cells). The number of contacts used along these bitlines depends on the technology and performance requirements. This type of array architecture is commonly referred to as a contact-less array because, unlike the conventional common source array architecture (wherein one contact is required for every two cells), the contact design rules do not limit the size of the cell. Therefore, scaling of the memory cell in such contact-less array architecture is made easier.
FIG. 1C is a circuit diagram of two rows and six columns of memory cells corresponding to the cross section and layout views in FIGS. 1A and 1B, respectively. This diagram shows the mirror image formation of the memory cells along each row, i.e., every two adjacent memory cells along a row are mirror images of one another.
The read, programming and erase operations of this array architecture are described in detail in the above-mentioned ""446 patent. Suffice it to state that programming is achieved through source side injection, and erasing is achieved through tunneling between the floating gate and the drain diffusion.
This flash EPROM approach possesses a number of drawbacks. First, during the deposition and definition of select gates 30 (FIGS. 1A and 1B), poly stringers form between adjacent rows of select gates 30, causing electrical shorts between them. The stringers form because the select gates 30 overlay a tall stack of first and second layer poly (approximately 4,000 xc3x85 high), and the conventional select gate etch, used in both the periphery and the array regions, does not fully remove the third layer poly in the array region, leaving behind poly stringers. Thus, additional etching in the array region is needed. Since the third layer poly in the periphery region does not require the over etching, an additional masking step is needed.
Second, the second layer poly (control gate 20) can not receive tungsten silicide (WSi2) due to the step height of the poly stack. Incorporating a tungsten silicide layer in the already tall stack of triple poly only exacerbates the problems associated with this stack, such as the stringers. However, without tungsten silicide, the RC time constant associated with control gates 20 is large, causing slow programming and erase functions.
Third, in high density memory devices, due to the typically large RC time delay associated with the polysilicon wordlines (select gates 30), strapping of the polysilicon wordlines with metal is required in order to achieve reasonable address access time. Such strapping requires drop contacts for making electrical contact between the poly wordline and the metal strap. The drop contacts result in larger array area.
Fourth, during the select gate oxidation step wherein the select gate oxide is formed, a phenomenon, commonly referred to as xe2x80x9ccuspingxe2x80x9d, occurs which results in a number of reliability problems. FIGS. 2A-2D illustrate this phenomenon. FIG. 2A shows the cross section of a stack of first layer polysilicon 10 (poly 1) and second layer polysilicon 20 (poly 2), the tunnel oxide 80 under poly 1, and the overlying layer of oxide 70. In FIG. 2B, the oxide layer 70 is removed through a dip off process, which as shown, results in removal of portions 81 of the tunnel oxide 80 under the outside edges of poly 1. In FIG. 2C, the gate oxidation step wherein gate oxide 90 is grown over the entire cell, results in raising of the outside edges of both poly 1 and poly 2. This phenomenon is commonly referred to as xe2x80x9csmiling polyxe2x80x9d. When the third layer of poly 30 (poly 3) is deposited over the gate oxide 90, as shown in FIG. 2D, the contours of poly 1 result in xe2x80x9ccuspingxe2x80x9d of poly 3 (i.e., poly 3 is pinched in the areas under the two ends of poly 1 as shown in the encircled region 82).
Cusping of poly 3 results in a number of reliability problems. First, the raised edges of poly 1 result in thicker tunnel oxide under these edges. This in turn results in slower erase since erase occurs through the tunnel oxide region between poly 1 and the drain diffusion 40 in the area marked as 82. Second, the oxide under the edges of poly 1 is formed from oxidized poly 1, which is a poor quality oxide. Such oxide possesses many trap sites which degrade the cycling characteristics of the device. Third, the cusping of poly 3 causes device failures due to charge loss during such reliability procedures as high voltage, high temperature dynamic burn-in cycles. Fourth, the cusping causes early retention failures during retention bake because the sharp corner of the cusp results in high fields.
In accordance with the present invention, a fully self-aligned, triple polysilicon, source side injection, nonvolatile memory cell suitable for use in a contact-less array of such cells wherein wordlines are overlaid with metal, as well as a method for producing the same is provided.
The following outlines one set of process steps for producing such contact-less array of nonvolatile memory cells in a silicon substrate: (a) a plurality of pairs of stacks of first and second layer polysilicon are formed along a row over the substrate; (b) a drain region is then formed in the substrate between the two stacks in each pair of polysilicon stacks, each drain region being self-aligned to the edges of the two stacks; (c) side-wall spacers are then formed adjacent to edges of each polysilicon stack; and (d) a source region is then formed in the substrate between each of two adjacent pairs of polysilicon stacks, the source region being self-aligned to the edges of the oxide spacers.
In one embodiment, a composite layer of, in the order from bottom to top, HTO-Nitride-Polysilicon (ONP) is formed over the array immediately after step (b). After step (d), the array surface is planarized using an insulating material, and then a trench is created over the row of cells by selectively removing the insulating material from over the row of cells. Next, the ONP composite layer is converted to an ONO composite layer, and the ONO composite layer is then anisotropically etched to form side-wall spacers adjacent to the edges of the polysilicon stacks. Select gate oxide is then grown over the row of cells, after which a third layer of polysilicon is formed over the select gate oxide. Finally, the third layer of polysilicon is overlaid with a layer of metal.
In yet another embodiment, a contact-less array of nonvolatile memory cells includes: a row of pairs of stacks of first and second polysilicon layers over a silicon substrate, the first polysilicon layer being insulated from the substrate, and the second polysilicon layer being insulated from the first polysilicon layer; a drain region in the substrate between the two stacks in each pair of polysilicon stacks, the drain region being aligned to the edges of the two stacks; a source region in the substrate between two adjacent pairs of polysilicon stacks, the source region being self-aligned to side-wall spacers formed adjacent to edges of the polysilicon stacks such that the source region is laterally spaced an equal distance from the edges of the two stacks of polysilicon between which the source region is located; and a third layer of polysilicon over but insulated from the row of polysilicon stacks and the silicon substrate, the third layer of polysilicon forming a wordline in the array.
One feature of the present invention is that the polysilicon layer in the ONP composite layer helps achieve a self-aligned source region by facilitating the use of oxide spacers in the source formation step.
Another feature is that a fully planarized array is obtained by utilizing the ONP composite layer in combination with the CMP technology, thereby eliminating all complications typical to non-planarized cell technologies, such as (i) the stringer problem discussed above and (ii) the inability to form tungsten polycide over the control gate due to the tall height of the polysilicon stack.
Yet another feature is that the xe2x80x9cONO spacersxe2x80x9d adjacent each stack of polysilicon prevent cusping of the third layer polysilicon thereby eliminating reliability problems associated with such cusping.
Yet another feature is that polysilicon wordlines are strapped with metal without use of drop contacts, thereby minimizing the wordline RC time delay without any area penalty.
Yet another feature is that the processing steps of the present invention can easily be integrated with conventional ETOX processes, thereby facilitating transporting these process steps to any manufacturing foundry using the ETOX process.
These and other features and advantages of the present invention will become more apparent from the following description and the accompanying drawings.